Problem 1

To explore the periodicity of the data, I folded the time so that it repeated every 90 mins which is the time to complete 1 orbit.

Similarly, there is periodicity in the particle counts by earth longitude because it repeats every 360 degrees and it takes 90 minutes to complete an orbit.

Here are some plots showing the periodic nature of the data by looking at the first 3 hours.

  1. The first plot shows a histogram of the particle count with a maximum of 30 particles. There is periodicity in the solar phase. It can be seen in the way the solar phase starts at 0 degrees, linearly increases until it gets to 360 degrees and then starts over at 0 degrees. The data also shows a 90 minute periodicity in Earth longitude where the longitude starts at 0, goes to 360 and starts over from 0. In the 'Particle Count by Earth Longitude' plot, there is an offset that occurs around 310 degrees. At greater than 310 degrees, there are more particles detected overall. This suggests that there is signal contamination for data from 310 degrees to 360 degrees. The background probability distribution can be constructed by looking at the distribution of particle counts binned by the other 3 column values.

  2. The background signal looks like an exponential decay from the range of -45 degrees to 315 degrees which occurs every 90 mins.

Because we are measuring particle counts, we know that the background distribution is going to be Poisson. The background data is time dependent so we need to make different background distributions for different times. To create a model of the background distribution, we need the average time between occurances which is determined by the most common particle count at specific times.

  1. I would say my model does not do a good job looking at the percent error plots which compare the model's probability to the data's probability. The model is more accurate when the particle count is close to the mode particle count. At the front and end tails of the distribution, there is over 100% error. The model also only takes into account the averaged probability of a particle count and not the overall probability.

  2. Using my model, I look at times 0, 10, 50, 80 mins. The 5$\sigma$ values for those times are 24, 22, 20, and 28 particles respectively.

Problem 2

Here is a plot of the images without any manipulation.

Since the images are all pointing at the same part of the sky but at different times, I subtracted one image from another to look at the background noise from the camera. This removes objects that are in both images like stars.

Next I plot the distribution of the noise on a semi-log y plot to confirm that it is Gaussian. The noise is the characteristic upside-down parabola which means it is Gaussian and it has mean of 0 and width of 1.

To find faint stars, I find the signal strength corresponding to 5$\sigma$ and filter out any signal less than that. This will leave me with images that only have a 5$\sigma$ or greater likelihood of being a star.

The plot above shows the stars in yellow. The images were filtered so that only signals stronger than 5$\sigma$ are shown.

Although we started with the same data, the background distributions are different because we used different methods of isolating the noise.